Wafer-level defect detection system and method for glass-based AM driving chip

Abstract

The invention discloses a wafer-level defect detection system and method for a glass-based AM driving chip, and relates to the technical field of chip detection. Bidirectional multispectral multi-angle irradiation is carried out on a to-be-detected area, and a first polarization spectrum vector set and a second polarization spectrum vector set of the to-be-detected area are obtained; performing reflection and transmission difference comparison of different polarization on the first polarization spectrum vector set and the second polarization spectrum vector set of the surface layer normal region, performing correlation characteristic analysis on each subsurface layer structure of the subsurface layer normal region, obtaining correlation windows of each subsurface layer structure, dividing the surface layer abnormal candidate region into a plurality of comparison windows, and performing comparison on the comparison windows; obtaining a personalized texture feature matrix of each comparison window, and marking the comparison window as a subsurface layer normal area or a subsurface layer abnormal area based on the personalized texture feature matrix of the comparison window; and the defect type corresponding to the abnormal area of the subsurface layer is obtained, so that the accuracy of wafer-level defect detection of the glass-based AM driving chip is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of chip detection, and specifically to a wafer-level defect detection system and method for glass-based AM driving chips. Background Art

[0002] The Chinese patent with the publication number CN115439427A discloses a wafer defect detection and positioning algorithm based on cascaded YOLO-GAN, including the following steps: during the manufacturing process of wafer production, the original image wafers are respectively fed into a wafer detection model based on improved YOLOv5 and a wafer semantic segmentation model based on BiseNet to obtain the positions of wafer target detection frames and the foreground masks of the wafers; the original images are input into a defect detection model based on an improved generative adversarial network to reconstruct the wafer images and locate the wafer defect regions; using the positions of the wafer target detection frames as constraints, the connected components of the defect images are analyzed, and a Softmax classifier is introduced to achieve the positioning of the defects and the subdivision of the wafer defects.

[0003] The Chinese patent with the publication number CN115112673A discloses a method and a detection system for detecting the appearance defects of wafer-level packaged chips, including the following steps: under the illumination of a high-angle annular light source and a low-angle annular light source, the camera separately images the chips on the tray, and according to a plurality of regional image templates, the corresponding regional images are respectively extracted from each chip image, and different second-level algorithms are used to process different regional images; the detection results of each second-level algorithm are summarized to obtain the detection result output of a single chip, and through cyclic detection, the detection results of each chip are obtained and summarized to obtain the result output of the entire tray of chips. Summary of the Invention

[0004] In order to solve the above technical problems, the purpose of the present invention is to provide a wafer-level defect detection method for glass-based AM driving chips, including the following steps: A wafer-level defect detection method for glass-based AM driving chips, including the following steps: Step s1: Perform bidirectional multi-spectral multi-angle illumination on the area to be detected, obtain the first polarization spectral vector set and the second polarization spectral vector set of the area to be detected, and perform regional feature correction; Step s2: Perform surface feature matching on the area to be detected, obtain the surface normal area or the surface abnormal area of the area to be detected, compare the first polarization spectral vector set and the second polarization spectral vector set of the surface normal area for the reflection and transmission differences of different polarizations, and mark the surface normal area as the subsurface normal area or the subsurface abnormal candidate area; Step s3: Conduct correlation feature analysis on each subsurface structure in the normal subsurface region, obtain the correlation window of each subsurface structure, construct the texture feature matrix of each correlation window, divide the surface abnormal candidate region into several comparison windows, obtain the personalized texture feature matrix of each comparison window according to the texture feature matrix of the correlation window, compare the texture feature matrix of the comparison window with the corresponding personalized texture feature matrix, and mark the comparison window as the normal subsurface region or the abnormal subsurface region; Step s4: Conduct abnormal feature analysis on the abnormal subsurface region and the surface abnormal region to obtain the defect types corresponding to the abnormal subsurface region and the surface abnormal region.

[0005] Further, the process of performing bidirectional multi-spectral multi-angle illumination on the region to be detected, obtaining the first polarization spectral vector set and the second polarization spectral vector set of the region to be detected and performing region feature correction includes: Preset feature acquisition parameters, perform bidirectional multi-spectral multi-angle illumination on the region to be detected according to the feature acquisition parameters, and obtain the reflected light image and the transmitted light image of the region to be detected; Extract features from each pixel point in the reflected light image to obtain the first polarization spectral vector set of each pixel point, and the first polarization spectral vector set includes the reflectance corresponding to different wavelengths under different polarization states; Extract features from each pixel point in the transmitted light image to obtain the second polarization spectral vector set of each pixel point, and the second polarization spectral vector set includes the transmittance corresponding to different wavelengths under different polarization states; Perform region feature correction on the first polarization spectral vector set and the second polarization spectral vector set.

[0006] Further, the process of performing region feature correction includes: Use a laser thickness gauge to detect the thickness of the region to be detected, obtain the actual thickness corresponding to each pixel point, obtain the standard thickness of the region to be detected, compare the actual thickness corresponding to each pixel point with the standard thickness to obtain the thickness deviation value, preset the thickness deviation threshold, and compare the thickness deviation value corresponding to the pixel point with the thickness deviation value; Construct a thickness compensation model. If the thickness deviation value corresponding to the pixel point is greater than the thickness deviation value, input the actual thickness corresponding to the pixel point into the thickness compensation model, and according to the thickness compensation model, output the theoretical reflectance and theoretical transmittance corresponding to different wavelengths under different polarization states of the pixel point. Adjust the reflectance corresponding to different wavelengths under the same polarization state of the first polarization spectral vector set according to the theoretical reflectance corresponding to different wavelengths under different polarization states, and adjust the transmittance corresponding to different wavelengths under the same polarization state of the second polarization spectral vector set according to the theoretical transmittance corresponding to different wavelengths under different polarization states.

[0007] Further, the process of performing surface feature matching on the area to be detected and obtaining the surface normal area or surface abnormal area of the area to be detected includes: Obtain the structures of each layer of the area to be detected and the penetration capabilities of different wavelengths for the structures of each layer, and set the sensitivity coefficients of different wavelengths for the structures of each layer according to the penetration capabilities of different wavelengths for the structures of each layer; Obtain the standard surface polarization spectrum vector set of each pixel point in the area to be detected. Based on the sensitivity coefficients of different wavelengths for the structures of each layer, perform surface feature matching between the first polarization spectrum vector set of the pixel point and the corresponding standard surface polarization spectrum vector set to obtain the surface feature similarity; Preset a surface feature similarity threshold. When the surface feature similarity is less than the surface feature similarity threshold, mark the pixel point as a surface abnormal area; When the surface feature similarity is greater than or equal to the surface feature similarity threshold, mark the pixel point as a surface normal area.

[0008] Among them, the calculation formula for obtaining the surface feature similarity is: ; Among them, is the surface feature similarity, is the sensitivity coefficient of the i-th wavelength for the surface structure, is the reflectance of the i-th wavelength in the j-th polarization state in the standard surface polarization spectrum vector set, represents the number of wavelengths, represents the number of polarization states.

[0009] Further, the process of comparing the reflection and transmission differences of different polarizations between the first polarization spectrum vector set and the second polarization spectrum vector set of the surface normal area and marking the surface normal area as a subsurface normal area or a subsurface abnormal area to be selected includes: Preset a defect-free threshold corresponding to the comprehensive polarization reflection and transmission difference index of the area to be detected. Compare the reflection and transmission differences of different polarizations between the first polarization spectrum vector set and the second polarization spectrum vector set of the surface normal area to obtain the comprehensive polarization reflection and transmission difference index; Compare the comprehensive polarization reflection and transmission difference index with the corresponding defect-free threshold. If the comprehensive polarization reflection and transmission difference index is greater than the defect-free threshold, mark the surface normal area as a subsurface abnormal area to be selected. If the comprehensive polarization reflection and transmission difference index is less than or equal to the defect-free threshold, mark the surface normal area as a subsurface normal area.

[0010] Among them, the calculation formula for obtaining the comprehensive polarization reflection and transmission difference index is: ; The background signal in the uniform area of the glass substrate can be effectively suppressed through PRDI, highlighting the polarization state deflection and spectral attenuation differences caused by defects.

[0011] Further, the process of obtaining the correlation window of each subsurface structure for the normal subsurface area includes performing a correlation feature analysis on each subsurface structure in the normal subsurface area: Select the current subsurface structure, randomly select a pixel from each pixel point marked as the normal subsurface area, mark the pixel as the starting pixel point, and perform the following steps: Step 1: Select the adjacent pixel points of the starting pixel point. According to the first polarization spectral vector sets of the starting pixel point and the adjacent pixel points, and the sensitivity coefficients of different wavelengths to the current subsurface structure, obtain the polarization spectral joint difference feature between the starting pixel point and the adjacent pixel points. Compare the polarization spectral joint difference feature with the preset polarization spectral joint difference feature threshold. If the polarization spectral joint difference feature is less than the polarization spectral joint difference feature threshold, mark the adjacent pixel point as the starting pixel point. If the polarization spectral joint difference feature is greater than or equal to the preset polarization spectral joint difference feature threshold, mark the adjacent pixel point as a non-correlated pixel point; Step 2: When the adjacent pixel points of the starting pixel point are marked as the starting pixel point, repeat Step 1 for the starting pixel point. When all the adjacent pixel points of the starting pixel point are non-correlated pixel points, mark the interval covered by all the starting pixel points as the correlation window of the current subsurface structure. Determine whether there are pixel points in the normal subsurface area of the current subsurface structure that are not covered by the correlation window. If there are, randomly select a pixel point from the pixel points not covered by the correlation window, mark the pixel point as the starting pixel point, and repeat Step 1 for the starting pixel point. If not, end the current step.

[0012] Further, the calculation formula for obtaining the polarization spectral joint difference feature between the starting pixel point and the adjacent pixel points is: Construct the polarization spectral vector of each pixel point for the current subsurface structure y according to the first polarization spectral vector set of each pixel point and the sensitivity coefficients of different wavelengths to the current subsurface structure ; ; is the sensitivity coefficient of the i-th wavelength to the current subsurface structure y; Construct the spectral angle mean of the starting pixel point u and the adjacent pixel point z according to the polarization spectral vector vy of each pixel point with respect to the sensitivity coefficient: ; where, represents the spectral angle mean of the starting pixel point u and the adjacent pixel point z, represents the polarization spectral vector of the starting pixel point u for the current surface structure y, Represents the polarization spectral vector of the adjacent pixel point z with respect to the current surface structure y; the smaller the PSA, the higher the local spectral polarization consistency, and the PSA of the regular texture area is close to 0; For each polarization direction , calculate the polarization difference index between the starting pixel point u and the adjacent pixel point z: ; ; Wherein, is the polarization difference between the starting pixel point u and the adjacent pixel point z, represents the maximum reflectance value corresponding to the polarization direction in the polarization spectral vector multiplied by represents the minimum reflectance value corresponding to the polarization direction in the polarization spectral vector multiplied by , represents the polarization difference corresponding to the polarization direction , represents the polarization difference corresponding to the polarization direction of the starting pixel point u, represents the polarization difference corresponding to the polarization direction of the adjacent pixel point z; According to the spectral angle mean value and the polarization difference index between the starting pixel point u and the adjacent pixel point z, obtain the polarization spectral joint difference feature of the starting pixel point u and the adjacent pixel point z : ; Wherein, represents the conversion coefficient. The above formulas are all calculated by removing the dimension and taking their numerical values. The formulas are obtained by software simulation of a large amount of data to get a formula closest to the actual situation. The preset parameters and preset thresholds in the formulas are set by those skilled in the art according to the actual situation or obtained by simulation of a large amount of data.

[0013] Furthermore, the process of constructing the texture feature matrix of each associated window includes: According to the sensitivity coefficients of different wavelengths to each subsurface structure and the first polarization spectral vector of each pixel point in each associated window of each subsurface structure, construct the polarization spectral vector of each pixel point in each associated window of each subsurface structure; Normalize the reflectance corresponding to different wavelengths under different polarization states in the polarization spectral vector of each pixel point within the associated window, quantize the normalized reflectance into discrete gray levels, determine the distance and direction of the gray-level co-occurrence matrix, and construct the gray-level co-occurrence matrix corresponding to different wavelengths under different polarization states based on the distance and direction of the gray-level co-occurrence matrix and the gray levels corresponding to different wavelengths under different polarization states; Obtain the texture features (contrast (CON), correlation (COR), energy (ENG), inverse difference moment (HOM)) corresponding to different wavelengths under different polarization states according to the gray-level co-occurrence matrix corresponding to different wavelengths under different polarization states, and construct the texture feature matrix of the associated window based on the texture features corresponding to different wavelengths under different polarization states , where, : ; where, represents the contrast corresponding to the polarization state and wavelength being and respectively; represents the correlation corresponding to the polarization state and wavelength being and respectively; represents the energy corresponding to the polarization state and wavelength being and respectively; represents the inverse difference moment corresponding to the polarization state and wavelength being and respectively.

[0014] Further, the process of dividing the surface anomaly candidate region into several comparison windows, obtaining the personalized texture feature matrix of each comparison window according to the texture feature matrix of the associated window, and comparing the texture feature matrix of the comparison window with the corresponding personalized texture feature matrix, and marking the comparison window as a subsurface normal region or a subsurface anomaly region includes: Obtain the polarization spectral vector of each pixel point within each subsurface structure in the subsurface anomaly candidate region according to the sensitivity coefficient of different wavelengths to each subsurface structure and the first polarization spectral vector of each pixel point in the subsurface anomaly candidate region; A preset comparison window, which is much smaller than the candidate area of the subsurface anomaly. The candidate area of the subsurface anomaly is split according to the comparison window into several comparison windows. For each pixel point in the comparison window, the reflectance corresponding to different wavelengths under different polarization states in the polarization spectral vector is normalized, and the normalized reflectance is quantized into discrete gray levels. The distance and direction of the gray-level co-occurrence matrix are determined. Based on the distance and direction of the gray-level co-occurrence matrix and the gray levels corresponding to different wavelengths under different polarization states, gray-level co-occurrence matrices corresponding to different wavelengths under different polarization states are constructed; according to the gray-level co-occurrence matrices corresponding to different wavelengths under different polarization states, texture features (contrast, correlation, energy, inverse difference moment) corresponding to different wavelengths under different polarization states are obtained, and according to the texture features corresponding to different wavelengths under different polarization states, a texture feature matrix of each comparison window within the subsurface structure is obtained; Obtain the texture feature matrices of each associated window within the subsurface structure, obtain the Euclidean distance between each comparison window and each associated window, set the weight coefficient of each associated window for each comparison window according to the Euclidean distance, and obtain the personalized texture feature matrix of each comparison window according to the weight coefficient of each associated window for each comparison window and the texture feature matrices of each associated window; The calculation formula for obtaining the personalized texture feature matrix of each comparison window is: ; Wherein, represents the weight coefficient of the associated window q for the comparison window p, represents the texture feature matrix of the associated window q, represents the total number of associated windows; Compare the texture feature matrix of each comparison window within the subsurface structure with the corresponding personalized texture feature matrix to obtain the texture feature similarity of each comparison window; A preset texture feature similarity threshold is set. If the texture feature similarity of the comparison window within the subsurface structure is less than the texture feature similarity threshold, the comparison window within the subsurface structure is marked as a subsurface anomaly area; if the texture feature similarity of the comparison window within the subsurface structure is greater than or equal to the texture feature similarity threshold, the comparison window within the subsurface structure is marked as a subsurface normal area.

[0015] Further, the process of performing abnormal feature analysis on the subsurface anomaly area to obtain the defect type corresponding to the subsurface anomaly area includes: Extract the difference features of the polarization spectral vector of each pixel point in the subsurface anomaly area within the subsurface structure to obtain the difference feature of each pixel point, and perform summation and averaging processing on the difference features of each pixel point to obtain the average difference feature; Among them, the differential features include the reflectivity difference of the same wavelength across polarization states (calculating the reflectivity difference of different polarization states (such as degrees, degrees)), and the same polarization across wavelengths (calculating the reflectivity gradient of adjacent wavelengths, such as ; A preset glass-based AM chip defect fingerprint library, which includes the average differential features corresponding to different defect types. Input the average differential features of the subsurface abnormal area into the glass-based AM chip defect fingerprint library for matching, obtain the similarity between different defect types in the subsurface structure and the subsurface abnormal area, preset a similarity threshold, and screen out the highest similarity from the similarities between different defect types and the subsurface abnormal area. If the highest similarity is greater than the similarity threshold, mark the defect type corresponding to the highest similarity in the subsurface abnormal area.

[0016] A wafer-level defect detection system for glass-based AM driver chips, including a monitoring center, which is communicatively connected to a data acquisition module, a data processing module, a data analysis module, and a defect annotation module; The data acquisition module is used to irradiate the area to be detected in two-way multi-spectral and multi-angle, obtain the first polarization spectral vector set and the second polarization spectral vector set of the area to be detected, and perform regional feature correction; The data processing module is used to perform surface feature matching on the area to be detected, obtain the surface normal area or the surface abnormal area of the area to be detected, compare the reflectance and transmittance differences of different polarizations between the first polarization spectral vector set and the second polarization spectral vector set of the surface normal area, and mark the surface normal area as the subsurface normal area or the subsurface abnormal candidate area; The data analysis module is used to perform correlation feature analysis on each subsurface structure of the subsurface normal area, obtain the correlation window of each subsurface structure, construct the texture feature matrix of each correlation window, divide the surface abnormal candidate area into several comparison windows, obtain the personalized texture feature matrix of each comparison window according to the texture feature matrix of the correlation window, compare the texture feature matrix of the comparison window with the corresponding personalized texture feature matrix, and mark the comparison window as the subsurface normal area or the subsurface abnormal area; The defect annotation module is used to perform abnormal feature analysis on the subsurface abnormal area and the surface abnormal area, and obtain the defect types corresponding to the subsurface abnormal area and the surface abnormal area.

[0017] Compared with the prior art, the beneficial effects of the present invention are: 1. Obtain rich spectral information, including the first polarization spectral vector set and the second polarization spectral vector set, through bidirectional multi-spectral multi-angle irradiation, and perform regional feature correction. Considering factors such as the thickness difference of different regions of the chip on the spectrum, the obtained spectral information can more accurately reflect the true state of the chip, thereby improving the recognition accuracy of defects.

[0018] 2. Considering the penetration ability, sensitivity coefficient, etc. of different wavelengths for each layer structure, it can adapt to the complex structure and material characteristics of glass-based AM-driven chips. Different chip structures and materials will have different optical responses under light of different wavelengths. By setting parameters such as sensitivity coefficients, targeted detection can be carried out according to the specific situation of the chip, which is applicable to wafer-level defect detection of various types and specifications of glass-based AM-driven chips.

[0019] 3. When analyzing the candidate areas with surface abnormalities, obtain the personalized texture feature matrix of each comparison window according to the texture feature matrix of the associated window. Fully considering the feature differences of different regions, the detection method is more flexible and adaptable. For example, for candidate areas with abnormalities in different positions, a personalized texture feature matrix can be constructed according to the texture features of the normal areas around them, so as to more accurately judge whether the area is a subsurface abnormal area, without being affected by the local feature differences of the chip.

[0020] 4. By analyzing the abnormal features of the subsurface abnormal areas and matching them with the preset defect fingerprint library of glass-based AM chips, the defect types corresponding to the subsurface abnormal areas can be accurately identified. The defect fingerprint library contains the average difference features corresponding to different defect types. By comparing the average difference features of the detected abnormal areas with the features in the fingerprint library, the defect types can be quickly and accurately determined. This helps subsequent targeted repair and quality control, improving the production quality and reliability of the chips. For example, when a subsurface abnormal area is detected in the chip, it can accurately judge which type of defect it is, such as material impurities, structural defects, etc., providing an accurate basis for subsequent process improvement and repair. Description of the Drawings

[0021] Figure 1 It is a schematic diagram of the wafer-level defect detection method for glass-based AM-driven chips according to an embodiment of the present application.

[0022] Figure 2 It is a schematic diagram of the wafer-level defect detection system for glass-based AM-driven chips according to an embodiment of the present application. Detailed Embodiments

[0023] Combined with the accompanying drawings in the embodiments of the present application, the technical solutions in the embodiments of the present application are clearly and completely described. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present application.

[0024] As Figure 1 shown, a wafer-level defect detection method for a glass-based AM driving chip includes the following steps: Step S1: Perform bidirectional multi-spectral multi-angle irradiation on the area to be detected, obtain the first polarization spectral vector set and the second polarization spectral vector set of the area to be detected, and perform regional feature correction; Step S2: Perform surface feature matching on the area to be detected, obtain the surface normal area or the surface abnormal area of the area to be detected, compare the first polarization spectral vector set and the second polarization spectral vector set of the surface normal area for reflection and transmission differences in different polarizations, and mark the surface normal area as the subsurface normal area or the subsurface abnormal candidate area; Step S3: Perform associated feature analysis on each subsurface structure of the subsurface normal area, obtain the associated window of each subsurface structure, construct the texture feature matrix of each associated window, divide the surface abnormal candidate area into several comparison windows, obtain the personalized texture feature matrix of each comparison window according to the texture feature matrix of the associated window, compare the texture feature matrix of the comparison window with the corresponding personalized texture feature matrix, and mark the comparison window as the subsurface normal area or the subsurface abnormal area; Step S4: Perform abnormal feature analysis on the subsurface abnormal area and the surface abnormal area to obtain the defect types corresponding to the subsurface abnormal area and the surface abnormal area.

[0025] It should be further noted that in the specific implementation process, the process of performing bidirectional multi-spectral multi-angle irradiation on the area to be detected, obtaining the first polarization spectral vector set and the second polarization spectral vector set of the area to be detected, and performing regional feature correction includes: Preset feature acquisition parameters, where the feature acquisition parameters include several polarization states (such as 4 polarization states: 0°, 45°, 90°, 135° linear polarization) and several wavelengths (such as visible light, infrared light, ultraviolet light, etc.), and perform bidirectional multi-spectral multi-angle irradiation on the area to be detected according to the feature acquisition parameters to obtain the reflected light image and the transmitted light image of the area to be detected; Extract features from each pixel point in the reflected light image to obtain the first polarization spectral vector set of each pixel point. The first polarization spectral vector set includes the reflectance corresponding to different wavelengths in different polarization states; Extract features for each pixel point in the transmitted light image to obtain the second polarization spectral vector set for each pixel point. The second polarization spectral vector set includes the transmittance corresponding to different wavelengths under different polarization states. Perform regional feature correction on the first polarization spectral vector set and the second polarization spectral vector set.

[0026] It should be further noted that in the specific implementation process, the process of performing regional feature correction includes: Use a laser thickness gauge to detect the thickness of the area to be detected, obtain the actual thickness corresponding to each pixel point, obtain the standard thickness of the area to be detected, compare the actual thickness corresponding to each pixel point with the standard thickness to obtain the thickness deviation value, preset the thickness deviation threshold, and compare the thickness deviation value corresponding to the pixel point with the thickness deviation value. Construct a thickness compensation model. If the thickness deviation value corresponding to the pixel point is greater than the thickness deviation value, input the actual thickness corresponding to the pixel point into the thickness compensation model. According to the thickness compensation model, output the theoretical reflectance and theoretical transmittance corresponding to different wavelengths under different polarization states of the pixel point. Adjust the reflectance corresponding to different wavelengths under the same polarization state of the first polarization spectral vector set according to the theoretical reflectance corresponding to different wavelengths under different polarization states, and adjust the transmittance corresponding to different wavelengths under the same polarization state of the second polarization spectral vector set according to the theoretical transmittance corresponding to different wavelengths under different polarization states.

[0027] Among them, the calculation formula of the thickness compensation model is: Let be the reflectance of the i-th wavelength in the j-th polarization state in the first polarization spectral vector set, represent the transmittance of the i-th wavelength in the j-th polarization state in the second polarization spectral vector set, be the actual thickness of the pixel point; ; ; ; ; Among them, represents the reflectance of the i-th wavelength in the j-th polarization state after regional feature correction in the first polarization spectral vector set, represents the transmittance of the i-th wavelength in the j-th polarization state after regional feature correction in the second polarization spectral vector set, represents the upper surface single reflection coefficient corresponding to the j-th polarization state, n represents the refractive index of the glass substrate, represents the incident angle, represents the refraction angle, determined by ​Denotes the i-th wavelength in the j-th polarization state.

[0028] It should be further noted that in the specific implementation process, the process of performing surface feature matching on the area to be detected and obtaining the surface normal area or surface abnormal area of the area to be detected includes: Obtain the layer structures of the area to be detected (including the surface structure and each subsurface structure) and the penetration capabilities of different wavelengths for each layer structure, and set the sensitivity coefficients of different wavelengths for each layer structure according to the penetration capabilities of different wavelengths for each layer structure; Obtain the standard surface polarization spectral vector set of each pixel point in the area to be detected. Based on the sensitivity coefficients of different wavelengths for each layer structure, perform surface feature matching between the first polarization spectral vector set of the pixel point and the corresponding standard surface polarization spectral vector set to obtain the surface feature similarity; Preset the surface feature similarity threshold. When the surface feature similarity is less than the surface feature similarity threshold, mark the pixel point as the surface abnormal area; When the surface feature similarity is greater than or equal to the surface feature similarity threshold, mark the pixel point as the surface normal area.

[0029] Among them, the calculation formula for obtaining the surface feature similarity is: ; Among them, is the surface feature similarity, is the sensitivity coefficient of the i-th wavelength for the surface structure, is the reflectivity of the i-th wavelength in the j-th polarization state in the standard surface polarization spectral vector set, represents the number of wavelengths, represents the number of polarization states.

[0030] It should be further noted that in the specific implementation process, the process of comparing the reflection and transmission differences of different polarizations between the first polarization spectral vector set and the second polarization spectral vector set of the surface normal area and marking the surface normal area as the subsurface normal area or the subsurface abnormal candidate area includes: Preset the defect-free threshold corresponding to the comprehensive polarization reflection and transmission difference index of the area to be detected. Compare the reflection and transmission differences of different polarizations between the first polarization spectral vector set and the second polarization spectral vector set of the surface normal area to obtain the comprehensive polarization reflection and transmission difference index; Compare the comprehensive polarization reflection and transmission difference index with the corresponding defect-free threshold. If the comprehensive polarization reflection and transmission difference index is greater than the defect-free threshold, mark the surface normal area as the subsurface abnormal candidate area. If the comprehensive polarization reflection and transmission difference index is less than or equal to the defect-free threshold, mark the surface normal area as the subsurface normal area.

[0031] Among them, the calculation formula for obtaining the comprehensive polarization reflection-transmission difference index is as follows: ; The background signal in the uniform area of the glass substrate can be effectively suppressed by the PRDI, highlighting the polarization state deflection and spectral attenuation differences caused by defects.

[0032] It should be further noted that in the specific implementation process, the process of performing correlation feature analysis on each subsurface structure in the normal subsurface area and obtaining the correlation window of each subsurface structure includes: Select the current subsurface structure, randomly select a pixel point from each pixel point marked as the normal subsurface area, mark the pixel point as the starting pixel point, and perform the following steps: Step 1: Select the adjacent pixel points of the starting pixel point. According to the first polarization spectral vector sets of the starting pixel point and the adjacent pixel points and the sensitivity coefficients of different wavelengths to the current subsurface structure, obtain the polarization spectral joint difference feature between the starting pixel point and the adjacent pixel points. Compare the polarization spectral joint difference feature with the preset polarization spectral joint difference feature threshold. If the polarization spectral joint difference feature is less than the polarization spectral joint difference feature threshold, mark the adjacent pixel point as the starting pixel point. If the polarization spectral joint difference feature is greater than or equal to the preset polarization spectral joint difference feature threshold, mark the adjacent pixel point as a non-correlated pixel point; Step 2: When the adjacent pixel points of the starting pixel point are marked as the starting pixel points, repeat Step 1 for the starting pixel point. When all the adjacent pixel points of the starting pixel point are non-correlated pixel points, mark the interval covered by all the starting pixel points as the correlation window of the current subsurface structure. Determine whether there are pixel points in the normal subsurface area of the current subsurface structure that are not covered by the correlation window. If so, randomly select a pixel point from the pixel points not covered by the correlation window, mark the pixel point as the starting pixel point, and repeat Step 1 for the starting pixel point. If not, end the current step.

[0033] It should be further noted that in the specific implementation process, the calculation formula for obtaining the polarization spectral joint difference feature between the starting pixel point and the adjacent pixel points is as follows: Construct the polarization spectral vector of each pixel point for the current subsurface structure y according to the first polarization spectral vector set of each pixel point and the sensitivity coefficients of different wavelengths to the current subsurface structure ; ; is the sensitivity coefficient of the i-th wavelength to the current subsurface structure y; Construct the spectral angle mean value between the starting pixel point u and the adjacent pixel point z according to the polarization spectral vector vy of each pixel point with respect to the sensitivity coefficient: ; Among them, represents the spectral angular mean value of the starting pixel point u and the adjacent pixel point z, represents the polarized spectral vector of the starting pixel point u with respect to the current surface structure y, represents the polarized spectral vector of the adjacent pixel point z with respect to the current surface structure y; the smaller the PSA, the higher the local spectral polarization consistency, and the PSA of the regular texture area is close to 0; For each polarization direction , calculate the polarization difference index between the starting pixel point u and the adjacent pixel point z: ; ; Among them, is the polarization difference between the starting pixel point u and the adjacent pixel point z, represents the maximum reflectance multiplied by corresponding to the polarization direction in the polarized spectral vector, represents the minimum reflectance multiplied by , represents the polarization difference corresponding to the polarization direction , represents the polarization difference corresponding to the polarization direction of the starting pixel point u, represents the polarization difference corresponding to the polarization direction of the adjacent pixel point z; According to the spectral angular mean value and the polarization difference index between the starting pixel point u and the adjacent pixel point z, obtain the polarization spectral joint difference feature : ; Among them, represents the conversion coefficient. The above formulas are all calculated by removing the dimension and taking their numerical values. The formulas are obtained by collecting a large amount of data for software simulation to get a formula closest to the real situation. The preset parameters and preset thresholds in the formulas are set by those skilled in the art according to the actual situation or obtained by a large amount of data simulation.

[0034] It should be further noted that in the specific implementation process, the process of constructing the texture feature matrix of each associated window includes: According to the sensitivity coefficient of each wavelength to each subsurface structure and the first polarized spectral vector of each pixel point in each associated window of each subsurface structure, construct the polarized spectral vector of each pixel point in each associated window of each subsurface structure; Normalize the reflectance corresponding to different wavelengths under different polarization states in the polarization spectral vector of each pixel in the associated window, quantize the normalized reflectance into discrete gray levels, determine the distance and direction of the gray-level co-occurrence matrix, and construct the gray-level co-occurrence matrix corresponding to different wavelengths under different polarization states based on the distance and direction of the gray-level co-occurrence matrix and the gray levels corresponding to different wavelengths under different polarization states; Obtain the texture features (contrast (CON), correlation (COR), energy (ENG), inverse difference moment (HOM)) corresponding to different wavelengths under different polarization states according to the gray-level co-occurrence matrix corresponding to different wavelengths under different polarization states, and construct the texture feature matrix of the associated window according to the texture features corresponding to different wavelengths under different polarization states , where :

[0035] where represents the contrast corresponding to the polarization state and wavelength being and respectively; represents the correlation corresponding to the polarization state and wavelength being and respectively; represents the energy corresponding to the polarization state and wavelength being and respectively; represents the inverse difference moment corresponding to the polarization state and wavelength being and respectively.

[0036] It should be further noted that in the specific implementation process, the surface anomaly candidate area is divided into several comparison windows, the personalized texture feature matrix of each comparison window is obtained according to the texture feature matrix of the associated window, and the process of comparing the texture feature matrix of the comparison window with the corresponding personalized texture feature matrix and marking the comparison window as a subsurface normal area or a subsurface anomaly area includes: Obtain the polarization spectral vector of each pixel in each subsurface structure in the subsurface anomaly candidate area according to the sensitivity coefficient of different wavelengths to each subsurface structure and the first polarization spectral vector of each pixel in the subsurface anomaly candidate area; A preset comparison window, which is much smaller than the candidate area of subsurface anomalies. The candidate area of subsurface anomalies is split according to the comparison window, divided into several comparison windows. The reflectances corresponding to different wavelengths under different polarization states in the polarization spectral vector of each pixel point within the comparison window are normalized, and the normalized reflectances are quantized into discrete gray levels. The distance and direction of the gray-level co-occurrence matrix are determined. Based on the distance and direction of the gray-level co-occurrence matrix and the gray levels corresponding to different wavelengths under different polarization states, gray-level co-occurrence matrices corresponding to different wavelengths under different polarization states are constructed; according to the gray-level co-occurrence matrices corresponding to different wavelengths under different polarization states, texture features (contrast, correlation, energy, inverse difference moment) corresponding to different wavelengths under different polarization states are obtained, and according to the texture features corresponding to different wavelengths under different polarization states, a texture feature matrix of each comparison window within the subsurface structure is obtained; Obtain the texture feature matrices of each associated window within the subsurface structure, obtain the Euclidean distance between each comparison window and each associated window (specifically, the Euclidean distance between the geometric center of the comparison window and the geometric center of the associated window), set the weight coefficient of each associated window for each comparison window according to the Euclidean distance, and obtain the personalized texture feature matrix of each comparison window according to the weight coefficient of each associated window for each comparison window and the texture feature matrices of each associated window; The calculation formula for obtaining the personalized texture feature matrix of each comparison window is: ; where, represents the weight coefficient of the associated window q for the comparison window p, represents the texture feature matrix of the associated window q, represents the total number of associated windows; Compare the texture feature matrix of each comparison window within the subsurface structure with the corresponding personalized texture feature matrix (calculated using cosine similarity) to obtain the texture feature similarity of each comparison window; A preset texture feature similarity threshold is set. If the texture feature similarity of the comparison window within the subsurface structure is less than the texture feature similarity threshold, the comparison window within the subsurface structure is marked as a subsurface anomaly area. If the texture feature similarity of the comparison window within the subsurface structure is greater than or equal to the texture feature similarity threshold, the comparison window within the subsurface structure is marked as a subsurface normal area.

[0037] It should be further noted that in the specific implementation process, the process of analyzing the abnormal features of the subsurface anomaly area and obtaining the defect type corresponding to the subsurface anomaly area includes: Extract the differential features of the polarization spectral vectors of each pixel point in the subsurface abnormal area within the subsurface structure, obtain the differential features of each pixel point, perform summation and averaging processing on the differential features of each pixel point, and obtain the average differential feature; Among them, the differential features include the reflectance difference across polarization states at the same wavelength (calculating the reflectance difference between different polarization states (such as degrees, degrees)), and the reflectance gradient across wavelengths at the same polarization (calculating the reflectance gradient between adjacent wavelengths, such as ; A preset glass-based AM chip defect fingerprint library, the glass-based AM chip defect fingerprint library includes the average differential features corresponding to different defect types, input the average differential feature of the subsurface abnormal area into the glass-based AM chip defect fingerprint library for matching (calculating the similarity between the average differential feature of the subsurface abnormal area and the average differential features corresponding to different defect types using cosine similarity), obtain the similarity between different defect types within the subsurface structure and the subsurface abnormal area, preset a similarity threshold, screen out the highest similarity from the similarities between different defect types and the subsurface abnormal area, if the highest similarity is greater than the similarity threshold, then mark the defect type corresponding to the highest similarity in the subsurface abnormal area; It should be further noted that in the specific implementation process, the process of performing abnormal feature analysis on the surface abnormal area and obtaining the defect type corresponding to the surface abnormal area includes: Extract the differential features of the polarization spectral vectors of each pixel point in the surface abnormal area within the surface structure, obtain the differential features of each pixel point, perform summation and averaging processing on the differential features of each pixel point, and obtain the average differential feature; Input the average differential feature of the surface abnormal area into the glass-based AM chip defect fingerprint library for matching, obtain the similarity between different defect types within the surface structure and the surface abnormal area, preset a similarity threshold, screen out the highest similarity from the similarities between different defect types and the surface abnormal area, if the highest similarity is greater than the similarity threshold, then mark the defect type corresponding to the highest similarity in the surface abnormal area.

[0038] As Figure 2 shown, a wafer-level defect detection system for a glass-based AM drive chip includes a monitoring center, and the monitoring center is communicatively connected to a data acquisition module, a data processing module, a data analysis module, and a defect annotation module; The data acquisition module is used to perform bidirectional multi-spectral multi-angle irradiation on the area to be detected, obtain the first polarization spectral vector set and the second polarization spectral vector set of the area to be detected and perform regional feature correction; The data processing module is used to perform surface feature matching on the area to be detected, obtain the surface normal area or surface abnormal area of the area to be detected, compare the first polarization spectrum vector set and the second polarization spectrum vector set of the surface normal area for reflection and transmission differences in different polarizations, and mark the surface normal area as the subsurface normal area or the subsurface abnormal candidate area; The data analysis module is used to perform correlation feature analysis on each subsurface structure of the subsurface normal area, obtain the correlation window of each subsurface structure, construct the texture feature matrix of each correlation window, divide the surface abnormal candidate area into several comparison windows, obtain the personalized texture feature matrix of each comparison window according to the texture feature matrix of the correlation window, compare the texture feature matrix of the comparison window with the corresponding personalized texture feature matrix, and mark the comparison window as the subsurface normal area or the subsurface abnormal area; The defect annotation module is used to perform abnormal feature analysis on the subsurface abnormal area and the surface abnormal area, and obtain the defect types corresponding to the subsurface abnormal area and the surface abnormal area.

[0039] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A method for wafer-level defect detection of a glass-based AM driving chip, characterized in that, Including the following steps: Step s1: Perform bidirectional multi-spectral multi-angle illumination on the area to be detected, obtain the first polarization spectral vector set and the second polarization spectral vector set of the area to be detected, and perform regional feature correction; Step s2: Perform surface feature matching on the area to be detected, obtain the surface normal area or surface abnormal area of the area to be detected, compare the first polarization spectral vector set and the second polarization spectral vector set of the surface normal area for reflection and transmission differences in different polarizations, and mark the surface normal area as the subsurface normal area or the subsurface abnormal candidate area; Step s3: Perform associated feature analysis on each subsurface structure of the subsurface normal area, obtain the associated window of each subsurface structure, construct the texture feature matrix of each associated window, divide the surface abnormal candidate area into several comparison windows, obtain the personalized texture feature matrix of each comparison window according to the texture feature matrix of the associated window, compare the texture feature matrix of the comparison window with the corresponding personalized texture feature matrix, and mark the comparison window as the subsurface normal area or the subsurface abnormal area; Step s4: Perform abnormal feature analysis on the subsurface abnormal area and the surface abnormal area, and obtain the defect types corresponding to the subsurface abnormal area and the surface abnormal area.

2. The method for wafer-level defect detection of a glass-based AM driving chip according to claim 1, characterized in that, The process of performing bidirectional multi-spectral multi-angle illumination on the area to be detected, obtaining the first polarization spectral vector set and the second polarization spectral vector set of the area to be detected, and performing regional feature correction Includes: Preset feature acquisition parameters, perform bidirectional multi-spectral multi-angle illumination on the area to be detected according to the feature acquisition parameters, and obtain the reflected light image and the transmitted light image of the area to be detected; Extract features from each pixel point in the reflected light image, obtain the first polarization spectral vector set of each pixel point, and the first polarization spectral vector set includes the reflectivities corresponding to different wavelengths in different polarization states; Extract features from each pixel point in the transmitted light image, obtain the second polarization spectral vector set of each pixel point, and the second polarization spectral vector set includes the transmittances corresponding to different wavelengths in different polarization states; Perform regional feature correction on the first polarization spectral vector set and the second polarization spectral vector set.

3. The method for wafer-level defect detection of a glass-based AM driving chip according to claim 2, characterized in that, The process of performing regional feature correction includes: Perform thickness detection on the area to be detected, obtain the actual thickness corresponding to each pixel point, obtain the standard thickness of the area to be detected, compare the actual thickness corresponding to each pixel point with the standard thickness, obtain the thickness deviation value, preset the thickness deviation threshold, and compare the thickness deviation value corresponding to the pixel point with the thickness deviation value; Construct a thickness compensation model. If the thickness deviation value corresponding to the pixel point is greater than the thickness deviation value, input the actual thickness corresponding to the pixel point into the thickness compensation model, output the theoretical reflectivity and theoretical transmittance corresponding to different wavelengths in different polarization states of the pixel point according to the thickness compensation model, adjust the reflectivities corresponding to different wavelengths in the same polarization state of the first polarization spectral vector set according to the theoretical reflectivity corresponding to different wavelengths in different polarization states, and adjust the transmittances corresponding to different wavelengths in the same polarization state of the second polarization spectral vector set according to the theoretical transmittance corresponding to different wavelengths in different polarization states.

4. The method for wafer-level defect detection of a glass-based AM driving chip according to claim 3, characterized in that, The process of performing surface feature matching on the area to be detected and obtaining the surface normal area or surface abnormal area of the area to be detected includes: Obtaining the structures of each layer of the area to be detected and the penetration capabilities of different wavelengths for the structures of each layer, and setting the sensitivity coefficients of different wavelengths for the structures of each layer according to the penetration capabilities of different wavelengths for the structures of each layer; Obtaining the standard surface polarization spectrum vector set of each pixel point in the area to be detected, constructing the polarization spectrum vector of each pixel point of the surface structure based on the sensitivity coefficients of different wavelengths for the structures of each layer and the first polarization spectrum vector of each pixel point, performing surface feature matching on the polarization spectrum vector set of each pixel point and the corresponding standard surface polarization spectrum vector set, and obtaining the surface feature similarity; Presetting a surface feature similarity threshold. When the surface feature similarity is less than the surface feature similarity threshold, mark the pixel point as a surface abnormal area; When the surface feature similarity is greater than or equal to the surface feature similarity threshold, mark the pixel point as a surface normal area.

5. The method for wafer-level defect detection of a glass-based AM driving chip according to claim 4, characterized in that, The process of comparing the reflection and transmission differences of different polarizations between the first polarization spectrum vector set and the second polarization spectrum vector set of the surface normal area and marking the surface normal area as a subsurface normal area or a subsurface abnormal area to be selected includes: Presetting a defect-free threshold corresponding to the comprehensive polarization reflection and transmission difference index of the area to be detected, comparing the reflection and transmission differences of different polarizations between the first polarization spectrum vector set and the second polarization spectrum vector set of the surface normal area, and obtaining the comprehensive polarization reflection and transmission difference index; Comparing the comprehensive polarization reflection and transmission difference index with the corresponding defect-free threshold. If the comprehensive polarization reflection and transmission difference index is greater than the defect-free threshold, mark the surface normal area as a subsurface abnormal area to be selected. If the comprehensive polarization reflection and transmission difference index is less than or equal to the defect-free threshold, mark the surface normal area as a subsurface normal area.

6. The method for wafer-level defect detection of a glass-based AM driving chip according to claim 5, characterized in that, The process of performing associated feature analysis on each subsurface structure of the subsurface normal area and obtaining the associated window of each subsurface structure includes: Select the current subsurface structure, randomly select a pixel point from each pixel point marked as a subsurface normal area, mark the pixel point as the starting pixel point, and perform the following steps: Step 1: Select the adjacent pixel points of the starting pixel point, obtain the polarization spectrum joint difference feature between the starting pixel point and the adjacent pixel points according to the first polarization spectrum vector sets of the starting pixel point and the adjacent pixel points and the sensitivity coefficient of the current subsurface structure for different wavelengths, compare the polarization spectrum joint difference feature with the preset polarization spectrum joint difference feature threshold. If the polarization spectrum joint difference feature is less than the polarization spectrum joint difference feature threshold, mark the adjacent pixel point as the starting pixel point. If the polarization spectrum joint difference feature is greater than or equal to the preset polarization spectrum joint difference feature threshold, mark the adjacent pixel point as a non-associated pixel point; Step 2: When the adjacent pixel points of the starting pixel point are marked as the starting pixel point, repeat Step 1 for the starting pixel point. When the adjacent pixel points of the starting pixel point are all non-associated pixel points, mark the intervals covered by all the starting pixel points as the associated window of the current subsurface structure. Determine whether there are pixel points in the normal subsurface area of the current subsurface structure that are not covered by the associated window. If so, randomly select a pixel point from the pixel points not covered by the associated window, mark the pixel point as the starting pixel point, and repeat Step 1 for the starting pixel point. If not, end the current step.

7. The method for wafer-level defect detection of a glass-based AM driving chip according to claim 6, characterized in that, The process of constructing the texture feature matrix of each associated window includes: Construct the polarization spectral vectors of each pixel point in each associated window of each subsurface structure according to the sensitivity coefficients of different wavelengths to each subsurface structure and the first polarization spectral vectors of each pixel point in each associated window of each subsurface structure; Normalize the reflectances corresponding to different wavelengths under different polarization states in the polarization spectral vector of each pixel point in the associated window, quantize the normalized reflectances into discrete gray levels, determine the distance and direction of the gray-level co-occurrence matrix, and construct the gray-level co-occurrence matrices corresponding to different wavelengths under different polarization states based on the distance and direction of the gray-level co-occurrence matrix and the gray levels corresponding to different wavelengths under different polarization states; Obtain the texture features corresponding to different wavelengths under different polarization states according to the gray-level co-occurrence matrices corresponding to different wavelengths under different polarization states, and construct the texture feature matrix of the associated window according to the texture features corresponding to different wavelengths under different polarization states.

8. The wafer-level defect detection method for a glass-based AM driving chip according to claim 7, characterized in that, The process of dividing the surface anomaly candidate area into several comparison windows, obtaining the personalized texture feature matrix of each comparison window according to the texture feature matrix of the associated window, and comparing the texture feature matrix of the comparison window with the corresponding personalized texture feature matrix and marking the comparison window as the normal subsurface area or the abnormal subsurface area includes: Obtain the polarization spectral vectors of each pixel point in each subsurface structure in the surface anomaly candidate area according to the sensitivity coefficients of different wavelengths to each subsurface structure and the first polarization spectral vectors of each pixel point in the surface anomaly candidate area; Preset comparison windows, split the surface anomaly candidate area according to the comparison windows into several comparison windows, and obtain the texture feature matrix of each comparison window in the subsurface structure based on the polarization spectral vectors of each pixel point in the subsurface structure of the surface anomaly candidate area; Obtain the texture feature matrices of each associated window in the subsurface structure, obtain the Euclidean distances between each comparison window and each associated window, set the weight coefficients of each associated window for each comparison window according to the Euclidean distances, and obtain the personalized texture feature matrix of each comparison window according to the weight coefficients of each associated window for each comparison window and the texture feature matrices of each associated window; Compare the texture feature matrix of each comparison window with the corresponding personalized texture feature matrix to obtain the texture feature similarity of each comparison window. Preset a texture feature similarity threshold. If the texture feature similarity of a comparison window within the subsurface structure is less than the texture feature similarity threshold, mark the comparison window as a subsurface abnormal area. If the texture feature similarity of a comparison window within the subsurface structure is greater than or equal to the texture feature similarity threshold, mark the comparison window as a subsurface normal area.

9. The wafer-level defect detection method for a glass-based AM driving chip according to claim 8, characterized in that, The process of performing abnormal feature analysis on the subsurface abnormal area to obtain the defect type corresponding to the subsurface abnormal area includes: Extract the differential features of the polarization spectral vectors of each pixel point in the subsurface abnormal area within the subsurface structure to obtain the differential features of each pixel point, and perform summation averaging on the differential features of each pixel point to obtain the average differential feature; Preset a defect fingerprint library for glass-based AM chips. The defect fingerprint library for glass-based AM chips includes the average differential features corresponding to different defect types of each layer structure. Input the average differential feature of the subsurface abnormal area into the defect fingerprint library for glass-based AM chips for matching to obtain the similarity between different defect types within the subsurface structure and the subsurface abnormal area. Preset a similarity threshold, and select the highest similarity from the similarities between different defect types and the subsurface abnormal area. If the highest similarity is greater than the similarity threshold, mark the defect type corresponding to the highest similarity in the subsurface abnormal area.

10. A wafer-level defect detection system for a glass-based AM driving chip, specifically applied to the wafer-level defect detection method for a glass-based AM driving chip according to any one of claims 1 to 9, characterized in that, It includes a monitoring center, and the monitoring center is communicatively connected to a data acquisition module, a data processing module, a data analysis module, and a defect annotation module; The data acquisition module is used to perform bidirectional multi-spectral multi-angle irradiation on the area to be detected, obtain the first polarization spectral vector set and the second polarization spectral vector set of the area to be detected, and perform regional feature correction; The data processing module is used to perform surface feature matching on the area to be detected to obtain the surface normal area or the surface abnormal area of the area to be detected, compare the first polarization spectral vector set and the second polarization spectral vector set of the surface normal area for reflection and transmission differences of different polarizations, and mark the surface normal area as a subsurface normal area or a subsurface abnormal candidate area; The data analysis module is used to perform correlation feature analysis on each subsurface structure of the subsurface normal area to obtain the correlation window of each subsurface structure, construct the texture feature matrix of each correlation window, divide the surface abnormal candidate area into several comparison windows, obtain the personalized texture feature matrix of each comparison window according to the texture feature matrix of the correlation window, compare the texture feature matrix of the comparison window with the corresponding personalized texture feature matrix, and mark the comparison window as a subsurface normal area or a subsurface abnormal area; The defect annotation module is used to perform abnormal feature analysis on the subsurface abnormal area and the surface abnormal area to obtain the defect types corresponding to the subsurface abnormal area and the surface abnormal area.

Images